RAVENZA MCP

Connect your AI to Ravenza

Ravenza exposes a hosted Model Context Protocol server. Your AI signs in through Ravenza and can build new AI Skills or access only the AI Skills enabled in your library.

Server URL

Codex

Register Ravenza, then complete its browser-based OAuth sign-in:

Ravenza opens a secure browser page where you can sign in or create an account with Google, GitHub, or email. Your AI should never ask for your password or one-time code. After you approve access, start a new Codex session if needed, then ask it to call search_packs to verify the connection.

Authentication

Use OAuth when your MCP client supports it. Ravenza publishes standard protected-resource metadata and uses Supabase Auth for OAuth 2.1 access tokens. Do not paste passwords or private keys into an AI client.

Available capabilities

Create a resumable Skill draft, add permitted sources, compile typed knowledge, run and store tests, keep the result private, or submit it for Ravenza review. Your connected AI performs the reasoning with its current model; Ravenza supplies the secure workflow, deterministic validation, storage, retrieval, and publishing controls. Ravenza does not call or bill a separate hosted model for creation.

The same connection can search and inspect AI Skills, manage your enabled library, search source-grounded knowledge, read source context, and compare evidence. Results include provenance and locator metadata for citation.

Guided creation

Tell your AI what the Skill should help with. It should ask one simple question at a time, handle Ravenza's technical fields itself, repair validation errors, and finish by asking: “Keep this private, or submit it to Ravenza?” Public submissions require your explicit confirmation of the title, description, cover, category, and source rights.

Agent discovery

AI agents and search systems can browse the complete public AI Skill directory, read its llms.txt guide, its full agent context, and the machine-readable discovery manifest, or fetch the public catalogue JSON. Individual AI Skill pages remain the canonical human-readable and indexable sources.

Ravenza's OAuth server is enabled. Compatible clients can discover authorization through the protected-resource metadata published beside the MCP endpoint.